Heenkenda, HMSCR & Fernando, TGI (2023). Chronological attribution of Sinhalese inscriptions using deep learning approaches. Journal of the National Science Foundation of Sri Lanka. https://jnsfsl.sljol.info/articles/10.4038/jnsfsr.v51i3.11200.
Abstract:
A study of this caliber can be identified as a profound source for a wealth of knowledge as the aim of this study is to present chronological attribution of Sinhalese inscriptions based on deep learning approaches. Inscriptions shed light on a multitude of information such as chronicled civilizational thought, economic status, language evolution, cultural boundaries, details of royal officers, local rules, ethnic groups, land tenure, religious activities, beliefs, and trade and industries. Inscriptions are major assets to showcase inclusive of listed above, multitude information; hence, the benefits served by a study of high caliber, especially to the historical heritage research and to the heritage tourism. Several computer-aided solutions have been proposed to resolve the recognition of inscriptions in the Sri Lankan context. But this paper proposes an optimized classification. A dataset of five hundred images of original Sinhalese inscriptions dating from the 3rd century BC to the present was used to train and test the models. This study adopts four deep learning models to classify Sinhalese inscriptions: a newly proposed convolutional neural network model, and the pre-trained models Inception-v3, VGG-19, and ResNet-50. Paleographical and morphological rules were adopted in the manual classification of Sinhalese inscriptions into a number of eras, namely, the Early Brahmi (3rd century BC to 1st century AD), Late Brahmi (2nd century AD to 4th century AD), Transitional Brahmi (5th century AD to 7th century AD), Medieval Sinhala (8th century AD to 14th century AD), and Modern Sinhala (15th century AD to the present). The results of the study indicate promising outcomes with accuracies of 70.66%, 85.94%, 57.44%, and 58.77% respectively for used four models. Further, the study revealed that the Inception-v3 model outperformed in classifying the Sinhalese inscriptions in respective eras.